[Pw_forum] TB09 metaGGA

Mon Nov 3 11:39:53 CET 2014

Dear Pascal

Please, have a look to these couple of papers:

http://dx.doi.org/10.1088/0953-8984/24/20/205503
http://dx.doi.org/10.1016/j.cpc.2013.02.020

They point to that Tran-Blaha modification to the Becke-Johnson metaGGA functional may yield quite good bandgaps for insulators (comparable to hybrid functionals) at a much reduced computational cost. The price to pay is that convergence is so tricky, even for simple binary compounds. I am giving them a chance to see if they deserve to be used instead of hybrid functionals. 

Thanks for your comments

Juanjo

From: Pascal BOULET 
Sent: Friday, October 31, 2014 3:17 PM
To: PWSCF Forum 
Subject: Re: [Pw_forum] TB09 metaGGA

Dear Juanjo,

As far as I know, metaGGAs are like GGAs for gaps: there are not accurate. For getting good gaps you have to switch to hybrid functionals or GW formalism (not implemented in QE though). Depending on what you want to do, of course, and if you know the gap value you may try this: optimize with a standard GGA functional, run a nscf calculation and if the gap is not totally closed up then operate a rigid band shifting for further properties calculations. This is a bit brute force...

If you do not know the gap value: run a hybrid single point calculation on the GGA optimized structure.

As Yves said, there are not so many options...

Pascal

"Juanjo Meléndez" <melendez at unex.es> wrote: 
  Dear Yves

  Thanks a lot for your comments. Unfortunately, I need a good description of gaps to calculate optical absorption, and hybrid functionals are prohibitive for the supercells that I need to use. That is why I started to try with metaGGA.

  I will have a look to the paper you mention (which I did not know, by the way, thanks!) for more ideas...

  Take care

  Juanjo

  From: Yves Ferro 
  Sent: Friday, October 31, 2014 10:20 AM
  To: PWSCF Forum 
  Subject: Re: [Pw_forum] TB09 metaGGA

  Dear Juanjo,  

  From what I can remember:
  - the meta-GGA kinetic terms are not saved in order to restart a nscf, at least up to the 5.0 version,
  - the tools for building meta-GGA PPs are not available in QE.
  However, I'm not sure using meta-GGA pseudos will solve your problem. Meta-GGA are known to have numerical instabilities such as the ones you met, in particular when then electronic density drop to a very low level as probably in your ionic crystal LiF, on the contrary with Si and C that form covalent structures.

  You can have a look to this paper you probably know:
  E.R. Johnson, JCP 131, 034111 2009

  I don't believe a cure to you problem exists at the moment, except by changing the functional to a GGA one.

  Best regards,

  Yves
  . 

  Le 31 oct. 2014 à 09:26, Juanjo Meléndez a écrit :

    Hi Pascal

    Thanks for replying. I agree with you that building a PP consistent with the metaGGA functional would solve most of the instability problems. But I do not have *any* experience building PPs (which, by the way, seems to be not trivial to me, even less trivial for metaGGA), so I followed http://dx.doi.org/10.1016/j.cpc.2013.02.020 and tried the trick. I was just wondering if any general strategy for convergence exists. 

    Regarding point 2, I asked one of the authors of the aforementioned paper and he declared not the have seen such a problem before, so I guess that the kinetic terms are not responsible. I have overcome the problem by using a dense kpoints path already in scf calculations and calculating the bands without the standard intermediate nscf run, but it would be nice if the developers could have a look and check. 

    Again, thank you very much for your interest.

    Juanjo

    Juan J. Meléndez 
    Associate Professor
    Department of Physics · University of Extremadura
    Avda. de Elvas, s/n 06006 Badajoz (Spain)
    Phone: +34 924 28 96 55
    Fax: +34 924 28 96 51
    Email: melendez at unex.es
    Web: http://materiales.unex.es/miembros/personal/jj-melendez/Index.html

    From: Pascal BOULET 
    Sent: Thursday, October 30, 2014 7:15 PM
    To: PWSCF Forum 
    Subject: Re: [Pw_forum] TB09 metaGGA

    hello, 

    I have never tried meta-GGA functionals with QE, so I am guessing. 

    For point 1, one possible reason is the inadequacy between the pseudo potential and the functional. I guess you have build a special pseudopotential for this functional.

    Point 2: Could it be because the meta-GGA kinetic terms are not saved in the checkpoint file and hence not readable when doing the nscf?

    Pascal

    "Juanjo Meléndez" <melendez at unex.es> wrote: 
      Hi all

      I am starting to work with metaGGA functionals in QE (v. 5.1 + libXc), following Èric Germaneau and colleagues’ work (http://dx.doi.org/10.1016/j.cpc.2013.02.020). I know that metaGGA is little tested, but I would be pleased if somebody could answer a couple of questions:

      1) Convergence is quite tricky. Up to know, I got some results in simple systems using a very small mixing_beta (within 0.01–0.05), not too many bands (just one or two conduction bands, never converged for more bands) and restarting from a previously fully converged calculation. I got results only for carbon, silicon and germanium, never got any convergence for any binary simple compound (like LiF). Does anyone have additional hints for metaGGA calculations to converge?

      2) In addition, I get something wrong with nscf calculations. After convergence of a scf run, I get a list of bands at each k-point, as usual –nothing strange here. Then I made a path of k-points to get the band structure and run a nscf calculation. This finishes fine as well, but the bands are not only different from those from the scf calculations, but also unrealistically high. I am attaching the input and output files for both the scf and nscf runs, as well as a couple of eigenval.xml files taken after nscf. Does anybody know how to manage this? Or could this be a bug in the code?

      Thanks a lot in advance

      Take care

      Juanjo

      Juan J. Meléndez 
      Associate Professor
      Department of Physics · University of Extremadura
      Avda. de Elvas, s/n 06006 Badajoz (Spain)
      Phone: +34 924 28 96 55
      Fax: +34 924 28 96 51
      Email: melendez at unex.es
      Web: http://materiales.unex.es/miembros/personal/jj-melendez/Index.html

      ****************************
      “All those who look stupid actually are, as also are half of those who do not look like” (F. de Quevedo)

      ****************************

--------------------------------------------------------------------------
              Este mensaje no contiene virus ni malware porque la protección de avast! Antivirus está activa. 

_______________________________________________Pw_forum mailing listPw_forum at pwscf.orghttp://pwscf.org/mailman/listinfo/pw_forum

    >-----------------
    Pascal Boulet
    Aix-Marseille University 
    MADIREL Laboratory
    Avenue Normandie-Niemen 
    13397 Marseille Cedex 20 
    Email: pascal.boulet at univ-amu.fr 
    Tel. +33 413 55 18 10 
    Fax +33 413 55 18 50

----------------------------------------------------------------------------
    _______________________________________________
    Pw_forum mailing list
    Pw_forum at pwscf.org
    http://pwscf.org/mailman/listinfo/pw_forum

----------------------------------------------------------------------------
            Este mensaje no contiene virus ni malware porque la protección de avast! Antivirus está activa. 

    _______________________________________________
    Pw_forum mailing list
    Pw_forum at pwscf.org
    http://pwscf.org/mailman/listinfo/pw_forum

------------------------------------------------------------------------------
  _______________________________________________
  Pw_forum mailing list
  Pw_forum at pwscf.org
  http://pwscf.org/mailman/listinfo/pw_forum

------------------------------------------------------------------------------
          Este mensaje no contiene virus ni malware porque la protección de avast! Antivirus está activa. 

_______________________________________________Pw_forum mailing listPw_forum at pwscf.orghttp://pwscf.org/mailman/listinfo/pw_forum

>-----------------
Pascal Boulet
Aix-Marseille University 
MADIREL Laboratory
Avenue Normandie-Niemen 
13397 Marseille Cedex 20 
Email: pascal.boulet at univ-amu.fr 
Tel. +33 413 55 18 10 
Fax +33 413 55 18 50

--------------------------------------------------------------------------------
_______________________________________________
Pw_forum mailing list
Pw_forum at pwscf.org
http://pwscf.org/mailman/listinfo/pw_forum

---
Este mensaje no contiene virus ni malware porque la protección de avast! Antivirus está activa.
http://www.avast.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20141103/658193f9/attachment.html>